4 ms·

> If you do something obvious like use amazon@example.com at amazon, you should expect them to add a rule that blah_blah_anything@example.com gets cleaned up to

> If you do something obvious like use amazon@example.com at amazon, you should expect them to add a rule that blah_blah_anything@example.com gets cleaned up to "@example.com" It seems like a very error prone thing to do automatically, while reaping little in return for the effort. How much do advertisers stand to gain by adding special case heuristics for "people who might be using an unshared email domain with a per-site local-part?" If I give my email as me+amazon@gmail.com, normalizing it to me@gmail.com is just applying a bit of common knowledge about how an email service with billion(s?) of users handles the + sign in the local-part of an address. It's a carve out, but it's totally deterministic (for Gmail at least) and applies to a lot of users. If I give my email as amazon@2d0d4809.com, normalizing it to @2d0d4809.com involves guesswork about that specific domain. I don't think normalizing away the + sign in the first case implies we should expect the latter to get normalized into a single identity as well.

fluidcruft 4y ago
I really don't know anything about the ad space but the linked project seems to indicate that ids get grouped into value classes. It describes IDs built off of normalized emails and also off of phone numbers to create identity-validated IDs which are of higher value in the ad market. Obviously phone numbers will be the most useful of the validated IDs and emails normalized. Of course if you have "@gmail.com" that's not going to have any value vs the identity-validated alternative. But at the bargain bin end of the ad selling universe does "amazon@2d0d4809.com" that has never been seen anywhere except amazon have more value than "@2d0d4809.com" that has been seen recently in a handful of other places consistent with use by a single person? It's obviously not an identity-validated ID class, so it's not going to demand identity-validated bids. But having a tracking history might add a tiny bit to the bid. So I wouldn't put it past people to use that rule.
prussian 4y ago
I don't even think the Heuristics would need to be that complex. Big email still uses reputation lists to determine if they should junk an email, so applying sender reputation, or lack thereof, to the domain part could be good enough to safely hash as an identity.

fluidcruft 4y ago

I really don't know anything about the ad space but the linked project seems to indicate that ids get grouped into value classes. It describes IDs built off of normalized emails and also off of phone numbers to create identity-validated IDs which are of higher value in the ad market. Obviously phone numbers will be the most useful of the validated IDs and emails normalized. Of course if you have "@gmail.com" that's not going to have any value vs the identity-validated alternative. But at the bargain bin end of the ad selling universe does "amazon@2d0d4809.com" that has never been seen anywhere except amazon have more value than "@2d0d4809.com" that has been seen recently in a handful of other places consistent with use by a single person? It's obviously not an identity-validated ID class, so it's not going to demand identity-validated bids. But having a tracking history might add a tiny bit to the bid. So I wouldn't put it past people to use that rule.

prussian 4y ago

I don't even think the Heuristics would need to be that complex. Big email still uses reputation lists to determine if they should junk an email, so applying sender reputation, or lack thereof, to the domain part could be good enough to safely hash as an identity.